Systems and methods for identifying subsurface hydrogen accumulation

Abstract

SOLUTION: Systems, apparatuses, methods and computer program products are disclosed for training an image analysis engine to identify surface features of the Earth consistent with subsurface hydrogen accumulation. An example method includes receiving, by communications circuitry, a training dataset of labeled images illustrating surface features consistent with subsurface hydrogen accumulation, where the surface features consistent with subsurface hydrogen accumulation include ovoid surficial depressions. The example method further includes training, by a model generator and using the training dataset, an image classification model of the image analysis engine to identify whether new images contain surface features consistent with subsurface hydrogen accumulation. The example method further includes hosting the image classification model by the image analysis engine.EFFECT: The present invention applies machine learning and remote sensing to automatically identify geomorphological features that are consistent with subsurface hydrogen accumulation.SELECTED DRAWING: Figure 1

Description

[Background technology]

[0001] Hydrogen is considered a fundamentally important resource in the anticipated global transition to a low-carbon future, and its demand is expected to increase significantly over the coming decades as society seeks fuel alternatives. While hydrogen is the most abundant element in the universe, today's primary sources of usable hydrogen are either carbon-intensive (e.g., steam-methane reforming or coal gasification) or energy-intensive (e.g., water electrolysis). As a result, most hydrogen is produced inefficiently, relegating it to the role of an energy carrier rather than a primary energy source. While large amounts of hydrogen are naturally produced below ground, subsurface exploration for natural hydrogen and related production strategies are still in their formative stages. Similar to the state of hydrocarbon exploration in the 1850s, almost all hydrogen drilling targets to date have been discovered serendipitously during exploration for petroleum, geothermal, or groundwater resources, or through the observation of hydrogen-rich surface seeps. Summary of the Invention

[0002] Discoveries of natural hydrogen (sometimes called geological or native hydrogen) have primarily been serendipitous or meandering discoveries, or based on observations of elevated hydrogen concentrations in surface seeps. Furthermore, the potential for hydrogen production in continental environments expands based on the abundance of suitable source rocks (mafic and ultramafic rocks account for more than 10% of the continental crust). However, exploration for geological reservoirs of natural hydrogen is significantly limited by an inadequate understanding of the conditions under which mafic and ultramafic rocks constitute a significant source of hydrogen; the formations that act as hydrogen reservoirs, traps, and seals; and the levels of hydrogen consumption in the shallow crust due to oxidation, biological availability, diffusional losses, and adsorption to clay minerals. To produce economical volumes of natural hydrogen, exploration strategies must be based on identifying suitable geological settings that enable hydrogen production and hydrogen retention, and prevent the permeation of oxidized or fresh water containing hydrogen-consuming microorganisms into the hydrogen reservoir. Once identified, drilling techniques used in both the petroleum and geothermal energy industries can be adapted to economically exploit natural hydrogen as a non-carbon source of energy or chemical feedstock.

[0003] Disclosed herein are methods, apparatus, and systems for identifying subsurface hydrogen accumulations. As described below, exemplary embodiments apply machine learning and remote sensing to automatically identify geomorphic features consistent with subsurface hydrogen accumulations. Furthermore, various exemplary embodiments describe techniques for analyzing such geomorphic features to identify those that are most likely to be active sources of producible hydrogen (e.g., sources currently producing hydrogen).

[0004] In one exemplary embodiment, a method is provided for training an image analysis engine to identify aboveground surface features consistent with subsurface hydrogen accumulation. The method includes receiving, via a communications circuit, a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation, where the surface features consistent with subsurface hydrogen accumulation include oval surface depressions; training, via a model generator and using the training dataset, an image classification model in the image analysis engine to identify whether images include surface features consistent with subsurface hydrogen accumulation; and hosting the trained image classification model by the image analysis engine.

[0005] In a related embodiment, a corresponding apparatus is provided for training an image analysis engine to identify aboveground surface features consistent with subsurface hydrogen accumulation, the apparatus including: a communications circuit configured to receive a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation, the surface features consistent with subsurface hydrogen accumulation including oval surface depressions; and a model generation unit configured to train an image classification model of the image analysis engine using the training dataset to identify whether images include surface features consistent with subsurface hydrogen accumulation, the image analysis engine being configured to host the trained image classification model.

[0006] In another related embodiment, a corresponding computer program product is provided for training an image analysis engine to identify aboveground surface features consistent with subsurface hydrogen accumulation. The computer program product includes at least one non-transitory computer-readable storage medium storing software instructions that, when executed, cause an apparatus to: receive a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation, the surface features consistent with subsurface hydrogen accumulation including oval surface depressions; and train an image classification model of the image analysis engine using the training dataset to identify whether images include surface features consistent with subsurface hydrogen accumulation, the trained image classification model being hosted by the image analysis engine.

[0007] In another exemplary embodiment, a method for automatically identifying aboveground surface features consistent with subsurface hydrogen accumulation is provided, the method including: receiving, by a communications circuitry, a target image; identifying, by an image analysis engine and using a trained image classification model, whether the target image includes any surface features consistent with subsurface hydrogen accumulation; and outputting, by the communications circuitry, an indication of whether the target image includes any surface features consistent with subsurface hydrogen accumulation.

[0008] In a related embodiment, a corresponding apparatus is provided for training an image analysis engine to identify aboveground surface features consistent with subsurface hydrogen accumulation, the apparatus including a communications circuit configured to receive a target image and an image analysis engine configured to identify whether the target image includes any surface features consistent with subsurface hydrogen accumulation using the trained image classification model, the communications circuit further configured to output an indication of whether the target image includes any surface features consistent with subsurface hydrogen accumulation.

[0009] In another related embodiment, a corresponding computer program product is provided for training an image analysis engine to identify aboveground surface features consistent with subsurface hydrogen accumulation. The computer program product includes at least one non-transitory computer-readable storage medium storing software instructions that, when executed, cause the device to receive a target image, use the trained image classification model to identify whether the target image includes any surface features consistent with subsurface hydrogen accumulation, and output an indication of whether the target image includes any surface features consistent with subsurface hydrogen accumulation.

[0010] In yet another illustrative embodiment, a method for automatically identifying subsurface surface features indicative of active subsurface hydrogen accumulation is provided, the method including: receiving, by a communications circuitry, information describing a subsurface oval-shaped surface feature; automatically estimating, by a relevance determination engine, a likelihood that the oval-shaped surface feature is indicative of active subsurface hydrogen accumulation; determining, by the relevance determination engine, whether the estimated likelihood meets a predetermined threshold; and, in instances where the estimated likelihood meets the predetermined threshold, outputting, by the communications circuitry, an indication that the surface feature is indicative of active subsurface hydrogen accumulation.

[0011] In yet another illustrative embodiment, a method for automatically identifying subsurface surface features indicative of active subsurface hydrogen accumulation is provided, the method including: receiving, by a communications circuitry, information describing a subsurface oval-shaped surface feature; automatically estimating, by a relevance determination engine, a likelihood that the oval-shaped surface feature is indicative of active subsurface hydrogen accumulation; determining, by the relevance determination engine, whether the estimated likelihood meets a predetermined threshold; and, in instances where the estimated likelihood meets the predetermined threshold, outputting, by the communications circuitry, an indication that the surface feature is indicative of active subsurface hydrogen accumulation.

[0012] In another related embodiment, a corresponding computer program product is provided for automatically identifying subsurface surface features indicative of active subsurface hydrogen accumulation. The computer program product includes at least one non-transitory computer-readable storage medium storing software instructions that, when executed, cause an apparatus to receive information describing a subsurface oval-shaped surface feature, automatically estimate a likelihood that the oval-shaped surface feature is indicative of active subsurface hydrogen accumulation, determine whether the estimated likelihood meets a predetermined threshold, and, in cases where the estimated likelihood meets the predetermined threshold, output an indication that the surface feature is indicative of active subsurface hydrogen accumulation.

[0013] Finally, yet another illustrative embodiment describes a method for automatically identifying terrestrial surface features indicative of active subsurface hydrogen accumulation. The method includes receiving, via a communications circuitry, a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation, and training, via a model generator and using the training dataset, an object detection model to identify regions within the images containing surface features consistent with subsurface hydrogen accumulation. The method further includes receiving, via the communications circuitry, target images, and identifying, via an image analysis engine and using the object detection model, regions within the target images containing surface features consistent with subsurface hydrogen accumulation. The method further includes automatically estimating, via a relevance determination engine, a likelihood that the identified surface features are indicative of active subsurface hydrogen accumulation, determining, via the relevance determination engine, whether the estimated likelihood meets a predetermined threshold, and outputting, via the communications circuitry, an indication that the surface features are indicative of active subsurface hydrogen accumulation.

[0014] In a related embodiment, a corresponding apparatus is provided for automatically identifying aboveground surface features indicative of active subsurface hydrogen accumulation. The apparatus includes a communications circuit configured to receive a training dataset of labeled images exhibiting surface features consistent with subsurface hydrogen accumulation; and a model generation apparatus configured to use the training dataset to train an object detection model to identify regions within the images containing surface features consistent with subsurface hydrogen accumulation, the communications circuitry further configured to receive target images. The apparatus further includes an image analysis engine configured to use the object detection model to identify regions within the target images containing surface features consistent with subsurface hydrogen accumulation, and a relevance determination engine configured to automatically estimate a likelihood that the identified surface features are indicative of active subsurface hydrogen accumulation and determine whether the estimated likelihood meets a predetermined threshold, the communications circuitry further configured to output an indication that the surface features are indicative of active subsurface hydrogen accumulation in cases where the estimated likelihood meets the predetermined threshold.

[0015] In another related embodiment, a corresponding computer program product is provided for automatically identifying aboveground surface features indicative of active subsurface hydrogen accumulation. The computer program product includes at least one non-transitory computer-readable storage medium having software instructions stored thereon that, when executed, cause an apparatus to receive a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation and use the training dataset to train an object detection model to identify regions within the images containing surface features consistent with subsurface hydrogen accumulation. The software instructions, when executed, further cause the apparatus to receive target images, identify regions within the target images containing surface features consistent with subsurface hydrogen accumulation, and automatically estimate a likelihood that the identified surface features are indicative of active subsurface hydrogen accumulation. The software instructions, when executed, further cause the apparatus to determine whether the estimated likelihood meets a predetermined threshold and, in cases where the estimated likelihood meets the predetermined threshold, output an indication that the surface features are indicative of active subsurface hydrogen accumulation.

[0016] The foregoing brief summary has been provided solely for the purpose of summarizing some exemplary embodiments described herein. The above-described embodiments are merely examples and should not be construed as narrowing the scope of the present disclosure in any way. It will be appreciated that the scope of the present disclosure encompasses many potential embodiments in addition to those summarized above, some of which are described in further detail below. [Brief explanation of the drawings]

[0017] Having described certain exemplary embodiments in general terms above, reference is now made to the accompanying drawings, which are not necessarily drawn to scale. Some embodiments may include fewer or more components than shown in the figures.

[0018] [Figure 1] 1 illustrates an exemplary system that some exemplary embodiments may use to identify underground accumulations of hydrogen. [Figure 2] 1 shows a schematic block diagram of an example circuit embodying a system device that may perform various operations in accordance with some example embodiments described herein. [Figure 3] This shows a Laser Imaging, Detection, and Positioning (LiDAR) elevation image of 300 square miles of Carolina Bay in Robeson County, North Carolina. [Figure 4] 1 illustrates an example flowchart for training an image analysis engine to identify above-ground surface features consistent with subsurface hydrogen accumulation, according to certain example embodiments described herein. [Figure 5] 1 illustrates an example flowchart for automatically identifying above-ground surface features consistent with subsurface hydrogen accumulation, according to certain example embodiments described herein. [Figure 6] 10 illustrates another example flowchart for automatically identifying above-ground surface features indicative of active subsurface hydrogen accumulation, according to certain example embodiments described herein. DETAILED DESCRIPTION OF THE INVENTION

[0019] Certain exemplary embodiments will now be described more fully hereinafter with reference to the accompanying drawings, in which some, but not necessarily all, embodiments are shown. Because the invention described herein may be embodied in many different forms, the invention should not be limited to only the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements.

[0020] The term "computing device" is used herein to refer to any one or all of industrial computers, desktop computers, personal data assistants (PDAs), laptop computers, tablet computers, smartbooks, personal computers, smartphones, wearable devices (e.g., headsets, smartwatches, or the like), programmable logic controllers (PLCs), programmable automation controllers (PACs), and similar electronic devices that include at least a processor and any other physical components necessary to perform the various operations described herein. Devices such as smartphones, laptop computers, tablet computers, and wearable devices are generally collectively referred to as mobile devices.

[0021] The terms "server" or "server device" are used to refer to any computing device that can function as a server, such as a master exchange server, a web server, a mail server, a document server, or any other type of server. A server may be a dedicated computing device or a server module (e.g., an application) hosted by a computing device that causes the computing device to act as a server.

[0022] overview Due to a lack of awareness of the existence of natural hydrogen accumulations in the subsurface, historically, there has not been a significant demand for natural hydrogen exploration techniques. As a result, there is a lack of knowledge regarding the existing volume of hydrogen accumulations in the subsurface, and there are no adequate conceptual models for identifying specific locations of subsurface accumulations of hydrogen. Therefore, identifying and evaluating subsurface natural hydrogen resources for extraction has been very difficult. Described herein are methods, apparatus, systems, and computer program products that enable the identification of topographic surface features that are likely to indicate the presence of active subsurface hydrogen accumulations.

[0023] The exemplary solution described herein is toS ource), move( M irrigation), reservoir ( R eservoir), trap ( T rap), and seal ( S In the context of a hydrogen SMaRTS system (Hydrogen SMaRTS) model, image classification tools are trained and applied that can automatically identify geographic features from satellite or airborne remote sensing observations as potential targets for natural subsurface hydrogen exploration. Further, through the application of machine learning algorithms (e.g., deep learning using neural networks) combined with analysis of remotely sensed data and other characteristics of the surface features of interest, exemplary embodiments enable the identification of potential natural hydrogen accumulations based on the proximity of suitable components of the hydrogen SMaRTS system to topographic surface features.

[0024] Underlying these solutions are new advances in understanding surface features consistent with subsurface hydrogen accumulation. In particular, ovoid surface features (also known as "bays" or "fairy circles") have been associated with hydrogen infiltration in the United States, Brazil, France, Russia, and elsewhere, and have recently been investigated for hydrogen production in Mali, Brazil, and the United States. Specifically, "Carolina Bays"-type ovoid features (shown in Figure 3), which include elliptical or circular surface depressions with diameters ranging from hundreds of meters to several kilometers, have been associated with focusing hydrogen migration pathways and hydrogen escape through hydrogen trap outflow points.

[0025] Known Carolina Bay ovoid surface features can be visually identified from freely available satellite imagery. Accordingly, exemplary embodiments described herein utilize machine learning techniques to detect these types of ovoids and distinguish those associated with surface hydrogen infiltration from those formed by other geomorphological processes (e.g., sinkholes, kettle flakes, metolitic impacts, etc.). Once hydrogen-related ovoid features (i.e., targets) are identified, imaging (e.g., LiDAR) and / or GPS monitoring of these features can be used to closely monitor surface deformation. Zones of active deformation are likely to be in close proximity (i.e., hydraulic communication) with subsurface systems currently experiencing natural hydrogen generation, migration, or water droplet reactions induced by hydrogen reactions during fluid migration and / or meandering. Conversely, static "target" (i.e., no deformation) zones are likely to record evidence of prior stages of hydrogen migration. In the former case, identification of active hydrogen systems can be used to prioritize drilling strategies and identify viable drilling targets when combined with robust characterization of hydrogen SMaRTS systems by conventional subsurface techniques (e.g., good petrophysical logging, sees reflectivity). In the latter case, based on subsequent characterization by conventional subsurface techniques (e.g., good petrophysical logging, sees reflectivity), this information can be used to eliminate some "targets" (preventing unnecessary drilling expenditures) or, alternatively, establish a secondary database of potential targets, depending on adequate characterization of adjacent hydrogen SMaRTS systems by conventional subsurface techniques (e.g., good petrophysical logging, sees reflectivity). Thus, satellite and airborne (e.g., LiDAR) data can identify previously unexplored targets and / or enable prioritization of zones of contemporary hydrogen activity, as evidenced by active surface deformation.

[0026] While many geographic "Carolina Bay-type" ovoid features could have been associated with hydrogen migration pathways at some point, only a (small) subset is associated with active hydrogen production systems. Furthermore, the subset of ovoid features associated with active hydrogen production may or may not contain the appropriate components of a hydrogen SMaRTS system to generate subsurface deposits of hydrogen or the presence of previously charged subsurface reservoirs of substantially recoverable hydrogen. However, the most attractive locations (i.e., highest priority targets) exhibit active / dynamic surface deformation (e.g., expansion or contraction) on the order of millimeters to centimeters per year, with the potential for major seasonal gyrations related to local precipitation patterns. The mechanisms driving the deformation may be multifold and include subsurface meandering and decarbonation reactions, which can be locally dilational (up to approximately 44% during CO2 mineralization into meanders, magnesite, and other carbonates).

[0027] Thus, exemplary embodiments are designed to detect dynamic ovoid features as promising targets for subsurface hydrogen exploration. Similar to landslides, fault zones, or sinkholes, actively deforming ovoid surface features exhibit increasingly divergent topographic and ecological characteristics from the surrounding environment. To detect such dynamics, high-resolution (<30 m) multispectral imagery can capture anomalies in vegetation and, in the case of substrates, changes that may be related to deformation rates. Similarly, the more active an ovoid is, the more its surface morphology should contrast with the surrounding environment. Thus, exemplary embodiments utilize digital elevation models (DEMs) with resolutions greater than 10 m horizontally and 5 m vertically in combination with multispectral imagery to identify these anomalous features. Finally, exemplary embodiments may, in some cases, directly reveal surface deformation via remotely sensed observations using LIDAR and interferometric synthetic aperture radar (InSAR) collected from both air and space, such as Sentinel-1, TerraSAR-X, and TanDEM-X. Successful measurements from InSAR require high microwave coherence between images and may therefore only be applicable to drier, low-vegetation areas. In any case, ovoid surface features suspected to be dynamic based on a combination of remotely sensed observations can then be subject to Global Navigation Satellite System (GNSS) surveys to provide geographic truth.

[0028] As described herein, exemplary embodiments provide methods and apparatus that enable improved identification of subsurface hydrogen accumulation through automated image analysis, remote sensing, and application of the SMaRTS conceptual model for hydrogen. Given the recent and expected future growth in demand for natural hydrogen combined with the carbon and / or energy intensity of current methods for producing man-made hydrogen, there is a large and growing need for tools that enable the production of natural hydrogen subsurface. The exemplary embodiments provide such tools that provide an automated, systematic, and comprehensive approach to identifying locations where hydrogen is likely to be produced, transported, and most importantly, stored (i.e., captured) within economical volumes subsurface.

[0029] While a high-level description of the operation of exemplary embodiments is provided above, specific details regarding the construction of some exemplary embodiments are provided below.

[0030] Underground hydrogen production In natural systems, the hydration of Fe-rich igneous rocks containing abundant olivine and pyroxene minerals is known to produce hydrogen, magnetite (Fe3O4), and other iron-bearing minerals through the serpentine reaction depicted below (Table 1). While most frequently observed along mesocosmic ridges where seawater interacts with heated mafic and ultramafic rocks, these reactions can also occur in continental environments where groundwater contacts iron-rich igneous intrusions, encompassing approximately 10% of the world's continental crust. Given the appropriate geographic setting, environmental conditions (e.g., temperature, pH, oxygen mobility, chemical composition, pressure), and water-rock interaction time, economical volumes of natural hydrogen can be produced and utilized as a carbon-free energy source. [Table 1]

[0031] However, subsurface exploration for natural hydrogen and related production strategies are in their formative stages and suffer from fundamental misunderstandings regarding the optimal geological setting and subsurface environmental conditions that will produce large quantities of natural hydrogen over time and adequately preserve it (i.e., prevent its consumption or conversion to another chemical phase). This allows natural hydrogen to accumulate and be stored in accessible and economical reservoirs underground. For example, the long-term persistence of stored hydrogen underground is compromised by its susceptibility to biodegradation or chemical oxidation even at relatively moderate temperatures (15-200°C), loss from reservoir fluids through clay adsorption or diffusive loss through sealed units, and / or consumption through non-pyrogenic methane formation at temperatures above approximately 200°C. While many of these factors are also relevant to oil and gas resources, the sensitivity to these conditions is significantly greater for the highly unstable hydrogen molecule.

[0032] Without a thorough understanding of these factors and the tools to predict their properties in the subsurface, developing exploration strategies and / or drilling programs to exploit natural hydrogen accumulations is inherently random, chaotic, and not economically or technically sound, as evidenced by the early history of petroleum exploration. Even today, despite significant technological advances in subsurface science, seep-drilling and exploratory drilling for hydrocarbon resources remains an extremely low-probability (and sometimes high-reward) endeavor. A thorough review of current hydrogen drilling strategies reveals that these same naive strategies are currently being deployed to explore for natural hydrogen. A significantly greater reliance on predictive subsurface properties and data science and data processing techniques is needed to enable the development of natural hydrogen exploration and production strategies based on hydrogen behavior.

[0033] SMaRTS model for hydrogen Petroleum geography utilizes the "SMaRTS" conceptual model, developed and refined over approximately 150 years of trial and error in petroleum exploration and production. The result of this very expensive, time-consuming, and iterative process is an approach for hydrocarbon generation, accumulation, and retention in sedimentary basements. These hydrocarbon "SMaRTS" systems are based on five key components (sources, reservoirs, and reservoirs) that, when present and with the appropriate characteristics, significantly improve the success of identifying and evaluating oil and gas resources in petroleum-deposited systems, as well as subsequent oil recovery. S ource), move( M irrigation), reservoir ( R eservoir), trap ( T rap), and seal ( S Includes

[0034] While various oil companies have their own versions and derivatives of the SMaRTS model, the general approach requires an understanding of: 1) source (e.g., hydrocarbon-based shale), which includes determining the rock units from which oil is generated, when and how much oil is generated; 2) migration, which includes the timing of oil migration from source rocks, how oil moves underground (i.e., its mechanism) (e.g., temporary in response to tectonic movements, continuous diffusion in response to leaking source rocks and seals), the extent to which oil is degraded as fluids flow to reservoirs, traps, and seals, and the later stages of hydrocarbon degradation and migration after initially reaching the reservoir, trap, or seal (i.e., tertiary migration); and 3) reservoir severities. These include: 1) dense characterization, which is a large, three-dimensional structure of porous, highly permeable formations (e.g., sandstone, limestone) ideal for accumulating and storing large amounts of hydrocarbons, allowing them to later flow relatively easily to production wells; 2) traps, which are structural or stratigraphic sequences surrounded by laterally extending impermeable units, establishing the three-dimensional volume of the resource accumulation; and 3) seals, which are impermeable units such as shale, evaporite, or low-porosity, unfractured (i.e., low-permeability) igneous bodies that prevent further suspension migration of hydrocarbons. Notably, the effectiveness of seals varies depending on the fluid composition. Characterizing the presence, integrity, and evolution of each of these components over time is crucial for economically generating resources from the hydrogen-bearing subsurface.

[0035] While SMaRTS strategies for hydrocarbon recovery are reliable and successful, no comparable conceptual model or workflow exists for natural hydrogen resource development. Furthermore, as noted above, initial strategies for hydrogen exploration models appear naive and immature, focusing on strategies relevant to hydrocarbon exploration rather than those appropriate for hydrogen resources. Thus, a need exists for the timely development of robust strategies for hydrogen SMaRTS. One such approach is described herein.

[0036] First, while both hard and soft rock analogs for layered / structural traps and high-quality seals targeted by the petroleum industry are suitable for natural hydrogen systems, the geological nature of these components can be very different. In particular, the geological setting, the formations that make up each of these components, and the interpretation of how they function over time often differ from traditional hydrocarbon exploration strategies, because these elements typically reside within or near igneous (mafic or ultramafic) bodies.

[0037] As a result, source rocks for economical natural hydrogen are mafic and ultramafic igniter rocks (containing olivine and pyroxene) and reduced iron (Fe 2+ ) content), as well as the metamorphic residues of these rocks. To locate natural hydrogen accumulations, it is important to locate hydrogen source rocks in geographic and layered regions with naturally optimized thermochemical conditions, hydraulic flow, and deformation. Because water is the reactant necessary for meandering to occur, sufficient hydraulic flow is required to contact the source rock surface area. Excessive primary or secondary porosity is rare in deep-buried lithologies; therefore, further exploration is needed to find source rocks with enhanced porosity, permeability, and / or brittle deformation (i.e., natural fractures) to achieve sufficient surface area for water droplet reaction to reach thermodynamic completion.

[0038] Characterizing the timing and mechanism of hydrogen transfer will also be useful for assessing the long-term stability of hydrogen and / or for evaluating the hydrogen SMaRTS model. RTS This is important for properly assessing formations adjacent to source rocks that are viable targets for components. In the case of hydrogen, fluid (e.g., hydrogen, pore water, geothermal water, and meteoric water introduced into the subsurface) migration is important given the potential for chemical oxidation, respiration, allogeneic methane formation, and / or diffusive loss. The mechanisms by which hydrogen migrates (e.g., drift) are similar to those for other deep-derived gases (e.g., CO2, hydrocarbons), including convective fluid flow or drift migration through porosity lithology and / or permeability deformation features (e.g., fractures, faults).

[0039] The optimal reservoir has high permeability and porosity and therefore may consist of sandstone or limestone, as in conventional hydrocarbon systems, when mafic and ultramafic rocks are emplaced in the sedimentary bed. However, given the possibility that hydrogen SMaRTS systems may exist in igneous and / or atypical terranes, fractured rock reservoirs are also likely to be an important reservoir type for hydrogen systems. It is important to note that the highly operative processes of igneous emplacement and / or meandering and mineralization associated with decarbonation may lead to strong fracturing of brittle crusts, potentially enabling the development of in situ fractured reservoirs.

[0040] For long-term retention of hydrogen, the reservoir must be within a trap covered by a suitable seal. Finally, a suitable seal must have very low porosity and permeability (i.e., nano-darcy scale) suitable to retard hydrogen flow, allowing hydrogen retention, while simultaneously preventing permeation of oxygenated or fresh water containing hydrogen-bearing microorganisms into the hydrogen reservoir.

[0041] Seals can consist of shale, evaporite, or other heavily cemented sedimentary units. However, the geographic setting of the expected hydrogen-producing reservoir also offers additional sealing options, such as superior low-porosity, unfractured sills, highly cemented / mineralized superior low-porosity, fractured sills, or evaporite deposited after rifting. In the latter case, natural meandering and decarbonation reactions may be sufficient to precipitate minerals in open pore spaces (pores, holes, and / or fractures), reducing hydraulic conductivity and creating self-sealing units that can prevent hydrogen generation. In the presence of these seals and reservoirs, three-dimensional structural or layered traps can be identified and accessed.

[0042] Common geographic settings where natural hydrogen production occurs predictably and has the potential for economic recovery include continental rifts (e.g., Mid-Continent Rift (USA), San Francisco Basin (Brazil), Triassic Rift Basins (East Coast)), ophiolites (e.g., New Caledonia (France), Coast Ranges Ophiolites (USA), Northern Oman), and areas where iron basaltic / tholeiitic volcanoes are found (e.g., Iceland, Salton Trough (USA and Mexico)).

[0043] System Architecture The exemplary embodiments described herein may be implemented using any of a variety of computing devices or servers. To this end, FIG. 1 illustrates an exemplary environment in which various embodiments may operate. As shown, the hydrogen targeting system 102 may include a system device 104 in communication with a data store 106. While the system device 104 and data store 106 are described in a single form, some embodiments may utilize two or more system devices 104 and / or two or more data stores 106. Additionally, some embodiments of the hydrogen targeting system 102 may not require a data store 106 at all and may instead access relevant geochemical or geophysical data from third-party data sources (not shown in FIG. 1 ) via a communications network 112 (e.g., the Internet) as needed. The hydrogen targeting system 102 further includes an image analysis engine 108, which may utilize machine learning modeling to analyze images and identify topographic features consistent with subsurface hydrogen accumulation. Hydrogen targeting system 102 further includes an associativity determination engine 110 that can estimate the likelihood that a given surface feature (identified via image analysis engine 108 or via another mechanism) indicates active subsurface hydrogen accumulation. As described in more detail below, image analysis engine 108 and associativity determination engine 110 can be components of system devices 104 or can be separate components of hydrogen targeting system 102. Regardless of implementation, hydrogen targeting system 102 and its constituent components can exchange information with any number of other devices, such as one or more remote sensing devices 114 and one or more user devices (e.g., user device 118A, user device 118B-user device 118N), via communications network 112.

[0044] System device 104 may be implemented as one or more servers, which may or may not be physically proximate to other components of hydrogen targeting system 102. Additionally, some components of system device 104 may be physically proximate to other components of hydrogen targeting system 102, but not to others. System device 104 may receive, process, generate, and transmit data, signals, and electronic information to facilitate operation of hydrogen targeting system 102. To this end, the memory of system device 104 stores control signals, device characteristics, and access credentials, enabling interaction between hydrogen targeting system 102 and one or more external devices, such as, for example, remote sensing device 114, user devices 118A-118N, etc. Certain components of system device 104 are described in more detail below with reference to apparatus 200 in connection with FIG. 2 .

[0045] The data store 106 may include components distinct from the system device 104 or may include elements of the system device 104 (e.g., memory 204, as described below in connection with FIG. 2). The data store 106 may be embodied as one or more direct-attached storage (DAS) devices (e.g., hard drives, solid-state drives, optical disk drives, etc.), or alternatively, may include one or more network-attached storage (NAS) devices independently connected to a communications network (e.g., communications network 112). The data store 106 may store information relied upon during operation of the hydrogen targeting system 102, such as geochemical datasets (e.g., fluid chemistry, well petrophysical records, seismic reflection data, etc.) for existing oil and geothermal wells and seeps, which may be publicly available through government agencies such as the Bureau of Land Management, the U.S. Geological Survey, the U.S. Department of Energy, or from external literature or proprietary sources of gas geochemical data. In this regard, the data store 106 may store an extensive collection of measurements of hydrogen and other important gases, and aqueous geochemical tracers (such as noble gases), from oil and gas, geothermal, CO2, and other industrial wells, humars, gas seeps, springs, and water-supply boreholes. The data store 106 may further store data on various stratigraphic units around the world, as well as seismic, gravity, or other geophysical data collected from various sources, such as the Advanced National Seismic System (ANSS), US Array, or other similar sources of extensive data on seismic activity around the world, and that may be used by the hydrogen targeting system 102.

[0046] Because the image analysis engine 108 and the relevance determination engine 110 may, in some embodiments, comprise components of the system device 104, these elements are described in more detail below in connection with the description of the various components of the system device 104. However, as described elsewhere herein, it will be understood that these components may comprise separate physical elements of the hydrogen targeting system 102, may be co-located with other components of the hydrogen targeting system 102, or, in some embodiments, may be located remotely from each other or from other components of the hydrogen targeting system 102.

[0047] The remote sensing device 114 may comprise any of a number of different remote sensing devices. For example, an exemplary remote sensing device 114 may be a satellite or satellite system such as either Landsat 7 or 8, or a space-borne device such as a higher-resolution commercial satellite such as DigitalGlobe / Maxar's WorldView constellation (satellites 1, 2, and 3). Additionally or alternatively, the remote sensing device 114 may include an airborne device such as an aircraft or unmanned aerial vehicle (UAV). Each remote sensing device 114 may acquire data about a particular topographic feature or area 116, either passively (where a reflection of another signal, such as sunlight, is captured) or actively (where the remote sensing device 114 emits a signal and detects the reflection of the signal). The captured data may include any of a variety of types of information, such as satellite or aerial observations (including pan-color, multispectral, or hyperspectral imagery), or surfaces (including vegetation, if present), or the elevation of the Earth's surface from photogrammetry, photochromometry, interferometry, LiDAR, or radar altimetry. )

[0048] One or more user devices 118A-118N may be embodied by any computing device known in the art, such as a desktop or laptop computer, a tablet device, a smartphone, or the like. User devices 118A-118N may be utilized by various individuals who interact with or operate hydrogen targeting system 102. For example, a first user may be in proximity to hydrogen targeting system 102 and utilize user device 118A to interact with hydrogen targeting system 102, while a second user may be located at the scene near a particular surface feature of interest and utilize user device 118B to interact with hydrogen targeting system 102. Any number of additional users may also utilize user devices to interact with hydrogen targeting system 102 or other users. One or more user devices 118A-118N need not be independent devices in themselves, but may be peripheral devices communicatively coupled to other computing devices.

[0049] 1 illustrates an environment and implementation in which hydrogen targeting system 102 interacts with any of user devices 118A-118N, in some embodiments, a user may interact directly with hydrogen targeting system 102 (e.g., via input / output circuitry of system device 104), in which case a separate user device may not be required. Through direct interaction or via a separate user device, a user may communicate with, operate, control, modify, or otherwise interact with hydrogen targeting system 102 to perform various functions and achieve various benefits described herein.

[0050] Exemplary Mounting Apparatus The system device 104 of the hydrogen targeting system 102 (described above with reference to FIG. 1) may be embodied by one or more computing devices or servers, shown as apparatus 200 in FIG. 2. As shown in FIG. 2, the apparatus 200 may include a processor 202, a memory 204, communications circuitry 206, input-output circuitry 208, a model generator 210, an image analysis engine 212, and a relevance determination engine 214, each of which is described in more detail below. Although the various components are only shown in FIG. 2 as being connected to the processor 202, the apparatus 200 may further include a bus (not explicitly shown in FIG. 2) for passing information among any combination of the various components of the apparatus 200. The apparatus 200 may be configured to perform the various operations described above with reference to FIG. 1 and below with reference to FIGS. 4-6.

[0051] The processor 202 (and / or coprocessors, or any other processors assisting or otherwise associated with the processor) may communicate with the memory 204 via a bus for passing information between components of the apparatus. The processor 202 may be embodied in several different ways, for example, may include one or more processing devices configured to be implemented independently. Furthermore, the processor may include one or more processors configured in tandem via a bus to enable independent execution of software instructions, pipelines, and / or multithreading. Use of the term “processor” may be understood to include a single core processor, a multicore processor, multiple processors in the apparatus 200, a remote processor or a “cloud” processor, or any combination thereof. As used herein, the term processor 202 may refer to any of several types of processing devices, including one or more central processing units (CPUs) generally designed to control the operation of the hydrogen targeting system 102, and, in particular, one or more separate graphics processing units (GPUs) that may be utilized by the model generation apparatus 210 and / or the image analysis engine 212 for training and utilizing various machine learning models.

[0052] Processor 202 may be configured to execute software instructions stored in, or otherwise accessible to, memory 204. In some cases, the processor may be configured to perform hard-coded functions. Thus, whether configured by hardware or software methods, or a combination of hardware and software, processor 202 represents an entity (e.g., physically embodied in circuitry) that can perform operations according to various embodiments of the present invention, while configured accordingly. Alternatively, as another example, when processor 202 is embodied as an executor of software instructions, the software instructions may specifically configure processor 202 to perform the algorithms and / or operations described herein when the software instructions are executed.

[0053] The memory 204 may be non-transitory and may include, for example, one or more volatile and / or non-volatile memories. In other words, for example, the memory 204 may be an electronic storage device (e.g., a computer-readable storage medium). The memory 204 may be configured to store information, data, content, applications, software instructions, etc. to enable the device to perform various functions in accordance with example embodiments contemplated herein. As previously mentioned, the data store 106 may be stored by the memory 204 in some embodiments.

[0054] Communications circuitry 206 may be any means, such as a device or circuitry embodied in either hardware or a combination of hardware and software, configured to receive and / or transmit data to and from a network and / or any other device, circuit, or module in communication with apparatus 200. In this regard, communications circuitry 206 may include, for example, a network interface for enabling communication with a wired or wireless communications network. For example, communications circuitry 206 may include one or more network interface cards, antennas, buses, switches, routers, modems, and supporting hardware and / or software, or any other devices suitable for enabling communication over a network. Additionally, communications circuitry 206 may include processing circuitry for causing the transmission of such signals to a network or for processing the reception of signals received from a network.

[0055] Apparatus 200 may include input-output circuitry 208 configured to provide output to a user and, in some embodiments, receive an indication of user input. Note that some embodiments will not include input-output circuitry 208, in which case user input may be received via a separate device, such as one of user devices 118A-118N (shown in FIG. 1). Input-output circuitry 208 may comprise a user interface, such as a display, and may further comprise components that manage use of the user interface, such as, for example, a web browser, a mobile application, a dedicated client device, etc. In some embodiments, input-output circuitry 208 may include a keyboard, a mouse, a touchscreen, a touch area, soft keys, a microphone, a speaker, and / or other input / output mechanisms. Input-output circuitry 208 may utilize processor 202 to control one or more functions of one or more of these user interface elements via software instructions (e.g., application software and / or system software such as firmware) stored in memory accessible to processor 202 (e.g., memory 204).

[0056] Additionally, the apparatus 200 further comprises a model generator 210 configured to use the training dataset to train an image classification model to identify whether an image contains images of topographic features consistent with subsurface hydrogen accumulation. As described in connection with FIG. 3 below, the model generator 210 may utilize the processor 202, the memory 204, or any other hardware components included in the apparatus 200 to perform these functions. The model generator 210 may be configured to train any of several different types of machine learning models, which may include image classification models. A particular image classification model trained by the model generator 210 may include an object detection model for identifying regions in an image containing surface features consistent with subsurface hydrogen accumulation, or a semantic segmentation model for identifying pixels of an image that correspond to surface features consistent with subsurface hydrogen accumulation. The model generator 210 may be configured to train any type of deep artificial neural network as an image classification model, such as a convolutional neural network (e.g., a residual neural network such as U-Net or ResNet). The model generator may train an image classification model as described below in connection with FIG. 4. The model generator 210 may further utilize communications circuitry 206 to transmit data to and / or receive data from various sources (e.g., user devices 118A-118N as shown in FIG. 1) and input-output circuitry 208 to transmit data to and / or receive data from users.

[0057] Additionally, apparatus 200 may also include an image analysis engine 212 configured to host a machine learning model trained to identify geomorphological features consistent with subsurface hydrogen accumulation. The machine learning model may comprise an image classification model that may be trained by model generator 210 to perform image analysis including image classification, object detection, semantic segmentation, etc. Image analysis engine 212 may utilize processor 202, memory 204, or any other hardware components included in apparatus 200 to perform its various operations, as described in connection with FIGS. 4 and 5 below. Image analysis engine 212 may further utilize communications circuitry 206 to collect data from various sources (e.g., user devices 118A-118N, data store 106, remote sensing device 114, or the like, as shown in FIG. 1 ) and may utilize input-output circuitry 208 to exchange data with a user. It should be understood that image analysis engine 212 may, in some embodiments, comprise a separate, dedicated element that enables its operation using dedicated physical components.

[0058] Finally, apparatus 200 may also include a relevance determination engine 214 configured to estimate the likelihood that a given oval surface feature indicates active subsurface hydrogen accumulation. Relevance determination engine 214 may utilize processor 202, memory 204, or any other hardware components included in apparatus 200 to perform this operation, as described in connection with FIG. 6 below. Relevance determination engine 214 may further utilize communications circuitry 206 to collect data from various sources (e.g., user devices 118A-118N, data store 106, remote sensing device 114, or the like, as shown in FIG. 1 ) and may utilize input-output circuitry 208 to exchange data with a user. It will be understood that relevance determination engine 214 may, in some embodiments, comprise a separate, dedicated element that enables its operation using dedicated physical components.

[0059] While components 202-214 are described in part using functional language, it will be understood that particular implementations necessarily include the use of specific hardware. It should also be understood that some of these components 202-214 may include similar or common hardware. For example, model generator 210, image analysis engine 212, and relevance determination engine 214 may each utilize the use of processor 202, memory 204, communications circuitry 206, or input-output circuitry 208, such that duplicated hardware is not required to facilitate the operation of these physical elements of device 200 (although in some embodiments, such as embodiments where increased parallelism may be desired, dedicated hardware elements may be used for any of these components). Thus, the use of the terms "circuitry" and "engine" with respect to elements of an apparatus should be interpreted as necessarily including specific hardware configured to perform the functionality associated with the particular element being described. Of course, the terms "circuitry" and "engine" should be understood broadly to include hardware, but in some embodiments, the terms "circuitry" and "engine" may also refer to software instructions that configure the hardware components of apparatus 200 to perform the various functions described herein.

[0060] While the model generator 210, the image analysis engine 212, and the relevance determination engine 214 may utilize the processor 202, memory 204, communications circuitry 206, or input-output circuitry 208 described above, it will be understood that any of these elements of the device 200 may include one or more dedicated processors, specially configured field programmable gate arrays (FPGAs), or application-specific interface circuits (ASICs) to perform the corresponding functions, and may therefore utilize the processor 202, or memory 204, communications circuitry 206, or input-output circuitry 208 executing software stored in a memory (e.g., memory 204) to enable any functionality not performed by a dedicated hardware element. However, it will be understood that in all embodiments, the model generator 210, the image analysis engine 212, and the relevance determination engine 214 are implemented via specific machines designed to perform the functions described herein with respect to such elements of the device 200.

[0061] In some embodiments, various components of the device 200 may be hosted remotely (e.g., by one or more cloud servers) and thus need not be physically present on the corresponding device 200. As previously mentioned, the image analysis engine 212 and the relevance determination engine 214 may comprise components separate from the system device 104 of the hydrogen targeting system 102. Thus, some or all of the functionality described herein may be provided by third-party circuitry. For example, a given device 200 may access one or more third-party circuitry via any type of network connection that facilitates the transmission of data and electronic information between the device 200 and the third-party circuitry. That device 200 may then remotely communicate with one or more of the other components described above to comprise the device 200.

[0062] As will be understood based on this disclosure, exemplary embodiments contemplated herein may be implemented by apparatus 200. Additionally, some exemplary embodiments may take the form of a computer program product including software instructions stored on at least one non-transitory computer-readable storage medium (e.g., memory 204). Any suitable non-transitory computer-readable storage medium may be utilized in such embodiments, some examples of which are non-transitory hard disks, CD-ROMs, flash memory, optical storage devices, and magnetic storage devices. It will be appreciated that, with respect to the particular device embodied by apparatus 200 as depicted in FIG. 2 , loading the software instructions into a computing device or apparatus creates a special-purpose machine comprising means for implementing the various functions described herein.

[0063] Having described specific components of the exemplary apparatus 200, the exemplary embodiment is described below in connection with a series of flowcharts.

[0064] Example Operation 4, 5, and 6, an exemplary flowchart is shown including exemplary operations associated with identifying subsurface accumulations of hydrogen. The operations illustrated in FIGS. 4-6 may be performed by a system device 104, which may be embodied, for example, by the hydrogen targeting system 102 shown in FIG. 1, and more specifically, by the apparatus 200 shown and described in connection with FIG. 2. To perform the operations described below, the apparatus 200 may utilize one or more of a processor 202, a memory 204, communication circuitry 206, input-output circuitry 208, a model generator 210, an image analysis engine 212, a relevance determination engine 214, and / or any combination thereof. It will be understood that user interaction with the hydrogen targeting system 102 may occur directly via the input-output circuitry 208 or may instead be facilitated by a separate user device (e.g., user devices 118A-118N as shown in FIG. 1), which may have similar or equivalent physical components to facilitate such user interaction.

[0065] Referring first to FIG. 4, an exemplary operation for training an image analysis engine to identify above-ground surface features consistent with subsurface hydrogen accumulation is shown.

[0066] As shown in operation 402, device 200 includes means, such as memory 204, communications circuitry 206, and input-output circuitry 208, for receiving a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation. The training dataset may be received from a separate device via communications circuitry 206. Additionally, or alternatively, the training dataset may be received from a user or peripheral device via input-output circuitry 208. Of course, in some scenarios, the training data may not be received from a separate source but may be stored locally on device 200. Alternatively, device 200 may acquire the data, such as by deploying it to a local system environment where the training data is already hosted, thereby reducing the computational burden of transmitting large amounts of training data from one location to another. As previously mentioned, received surface features consistent with subsurface hydrogen accumulation may include oval surface depressions. Of course, other surface features are shown in the various received images that are consistent with the presence of subsurface hydrogen accumulation and, in some cases, may separately indicate the presence of subsurface hydrogen accumulation. In implementations where the image classification model to be trained includes an object detection model, the training dataset may also include a set of bounding boxes for each image in the set of labeled images, where each bounding box for a particular image encloses a corresponding segment of the image that includes surface features consistent with subsurface hydrogen accumulation. In implementations where the image classification model to be trained includes a semantic segmentation model, the training dataset may include a set of pixel-wise masks for each image in the set of labeled images, where the set of pixel-wise masks for a particular image collectively identify whether all pixels of the particular image correspond to surface features consistent with subsurface hydrogen accumulation.

[0067] As indicated by operation 404, the apparatus 200 includes a means, such as a model generator 210, for training an image classification model using the training dataset. The model generator 210 may be configured to train an artificial neural network as the image classification model. The artificial neural network may include a fully convolutional neural network. Fully convolutional neural networks, such as those in the "U-Net" family, include several "decoders" and "encoders," or downsamplers and upsamplers, that learn to identify features at multiple spatial scales. A drawback of readily available U-Net implementations is that they often require fixed-size input images and large amounts of computer memory, even for small images. Because images received from the remote sensing device 114 are often significantly larger than images typically utilized by such modeling solutions, the use of deep neural networks with many hidden layers, such as U-Net and ResNet, may exhibit certain inefficiencies. Accordingly, the model generator 210 may also utilize modified versions of convolutional neural networks that enable direct processing of full-size Landsat, GeoEye, QuickBird, WorldView, Planet, or other satellite imagery (i.e., approximately 1 billion pixels) on consumer graphics cards (GPUs) within seconds. Such modifications may involve any of the following: 1) an input layer that allows for arbitrary image sizes; 2) reducing the number or encoder / decoder layers in U-Net-like networks; 3) optimizing kernel size within each convolutional layer; 4) using strides instead of max-pooling to spatially downscale inputs; 5) increasing the receiving field of the neural network; and / or 6) keeping the size of feature maps constant across all layers of the neural network. These optimizations allow for a large number of trainable parameters while minimizing memory and computational requirements.

[0068] In some embodiments, the image classification model includes an object detection model, in which case the model generator 210 may train the image classification model to identify regions in the image that contain surface features consistent with subsurface hydrogen accumulation. In other embodiments, the image classification model includes a semantic segmentation model, in which case the model generator 210 may train the image classification model to identify all pixels in the image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0069] As indicated by operation 406, the apparatus 200 includes means for hosting the trained image classification model, such as, for example, the image analysis engine 212. Hosting the image classification model may include storing the trained model in memory associated with the image analysis engine 212 to facilitate subsequent use of the trained image classification model by the image analysis engine 212.

[0070] In some embodiments, the procedure may end after completion of operation 406. However, in other embodiments, the procedure may then proceed to the set of operations shown in Figure 5 and described below, in which case the trained image classification model may be applied to analyze new images received by device 200.

[0071] Referring now to Figure 5, exemplary operations for automatically identifying surface features consistent with subsurface hydrogen accumulation are shown. As noted above, the sequence of operations shown in Figure 5 may be performed as an initial set of operations in a given procedure, or may be performed after training an image classification model such as that described in Figure 4, in which case operation 502 may be reached after completion of operation 406.

[0072] As indicated by operation 502, apparatus 200 includes means for receiving an image of interest, such as memory 204, communications circuitry 206, and input-output circuitry 208. In some cases, the image of interest may be a panchromatic, multispectral, or hyperspectral image, a satellite image, a panchromatic, multispectral, or hyperspectral satellite image, or any other type of image that may capture surface features. The image of interest may be received from a variety of sources. For example, the image of interest may be received from local memory 204 of apparatus 200, which may have previously stored the image of interest after it was received by apparatus 200 from a separate device. The image of interest may alternatively be received by communications circuitry 206, which may receive the image of interest from a separate device, such as remote sensing device 114, a user device (e.g., one of user devices 118A-118N), or a remote data store containing the image of interest. Still further, information may be received from input-output circuitry 208 in scenarios in which the image of interest is provided directly by a user, such as via a peripheral device.

[0073] As indicated by operation 504, apparatus 200 includes means, such as image analysis engine 212, for analyzing the target image to determine the presence of surface features consistent with subsurface accumulation of hydrogen. Before analyzing the target image, image analysis engine 212 may, in some cases, first perform orthorectification on the target image to remove distortions caused by the relative positions of the satellites capturing the image and the surface features depicted in the target image. Orthorectification may not be necessary in all embodiments, and the need for orthorectification of the target image may depend on the source of the target image. To analyze the target image, image analysis engine 212 may use a trained image classification model to identify whether the target image contains any surface features consistent with subsurface hydrogen accumulation. This trained image classification model may include a deep neural network, as described above. In some embodiments, the image classification model may include an object detection model, in which case image analysis engine 212 may use the object detection model to analyze the target image to identify all regions within the target image that contain surface features consistent with subsurface hydrogen accumulation. In some embodiments, the image classification model may include a semantic segmentation model, in which case the image analysis engine 212 may analyze the target image using the semantic segmentation model to identify all pixels in the target image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0074] It will be appreciated that the image analysis engine 212 is not limited to analyzing a single image, but may instead analyze a sequence of target images. To this end, the image analysis engine 212 may receive a set of target images and analyze all of the target images in the set of target images.

[0075] As indicated by operation 506, the device 200 includes means, such as the memory 204, the communications circuitry 206, the input-output circuitry 208, and the image analysis engine 212, for outputting the results of the analysis of the target image. This output may identify whether the target image includes any surface features consistent with subsurface hydrogen accumulation. In implementations in which the image classification model includes an object detection model, the output may identify any regions in the target image that include surface features consistent with subsurface hydrogen accumulation, such as by overlaying bounding boxes on the image, each bounding box including a corresponding associated region. In implementations in which the image classification model includes a semantic segmentation model, the output may identify all pixels in the target image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0076] If the apparatus 200 analyzes multiple target images, the apparatus 200 may then output an indication of all target images of the multiple target images that contain surface features consistent with subsurface hydrogen accumulation. To this end, the apparatus 200 may include a means, such as the relevance determination engine 214, for clustering various identified surface features together based on their similarity. To do this, the model generation unit 210 may train a supervised or unsupervised machine learning model to perform clustering on the surface features identified in the above-described manner, using K-means / K-median clustering, mean-shift clustering, or density-based spatial clustering of applications with noise (DBSCAN), Gaussian mixtures, or the like. After identifying the set of surface features, the apparatus 200 may include a means, such as the image analysis engine 212, for clustering the set of surface features using the trained clustering model. The relevance determination engine 214 may then utilize the clustered surface features to adjust weights applied to the likelihood that various surface features indicate active subsurface hydrogen accumulation (the estimation of which is described below in connection with operation 604 of FIG. 6 ).

[0077] In some embodiments, the procedure may end after completing operation 506. However, in other embodiments, the procedure may then proceed to a series of operations illustrated in Figure 6 and described below, in which the identified surface features are analyzed to estimate their likelihood of indicating active subsurface hydrogen accumulation.

[0078] Referring now to Figure 6, exemplary operations for automatically identifying whether surface features indicate active subsurface hydrogen accumulation are shown. As directly noted above, the sequence of operations shown in Figure 6 may be performed as an initial set of operations in a given procedure, or may be performed after analysis of one or more target images by apparatus 200 as described in Figure 5, in which case operation 602 may be reached after completion of operation 506.

[0079] As indicated by operation 602, apparatus 200 includes means, such as processor 202, memory 204, communications circuitry 206, input-output circuitry 208, model generator 210, image analysis engine 212, and relevance determination engine 214, for receiving information describing subsurface oval surface features. This information may be received from a variety of sources. For example, information may be received from image analysis engine 212 upon identification by image analysis engine 212 of a target image containing surface features consistent with subsurface hydrogen migration to the surface. As previously mentioned, such surface features may often include oval surface features. Alternatively, this information may be received from local memory 204 of apparatus 200, which may have previously stored the information during performance of the procedure described in connection with FIG. 5 or which may have previously received the information from another device. 5, or from a user device (e.g., one of user devices 118A-118N), or from a separate data store containing information describing oval surface features. Still further, the information may be received from input-output circuitry 208 in scenarios where the information is provided directly by a user, such as via a peripheral device.

[0080] As mentioned above, ovoid surface features, particularly surface depressions, are sometimes caused by hydrogen infiltration from the subsurface and thus comprise geomorphological features consistent with subsurface hydrogen accumulation. However, many ovoid surface features, despite being geomorphological features consistent with subsurface hydrogen accumulation, are created by geological events unrelated to hydrogen accumulation (e.g., sinkholes, kettle flakes, metolite impacts, etc.). In addition, many ovoid surface features are not connected to active hydrogen accumulation and were created by hydrogen infiltration from depleted systems. Therefore, further evaluation of a given ovoid feature allows for the determination of whether it is likely to be an indicator of active subsurface hydrogen accumulation.

[0081] While on-site evaluation can provide a solution for determining whether an oval surface feature is caused by active hydrogen infiltration from the subsurface, there are thousands of oval surface features worldwide, and on-site evaluation of all such surface features is impractical. Thus, a need exists to further examine known oval surface features with tools that can estimate the likelihood that a given oval surface feature is an indicator of active subsurface hydrogen accumulation. In doing so, oval surface features can be identified and ranked for their relevance to active subsurface hydrogen accumulation, thereby enabling the assignment of on-site exploration solutions.

[0082] As indicated by operation 604, the apparatus 200 includes means, such as the relevance determination engine 214, for automatically estimating the likelihood that a given oval surface feature indicates active subsurface hydrogen accumulation. Various methods for automatic likelihood estimation are contemplated herein, which may generally be categorized into four categories. The first category includes detecting indicators of geomorphic differentiation between the oval surface feature and its surrounding topography. The second category includes detecting indicators of the rate of change of one or more contours of the oval surface feature. The third category includes detecting indicators related to the stratigraphic unit containing the oval surface feature. And, the fourth category includes detecting indicators that may be evident from the specific contours of the oval feature. Thus, the relevance determination engine 214 may estimate the likelihood that an oval surface feature indicates active subsurface hydrogen accumulation based on (i) the degree of geomorphic differentiation between the oval surface and one or more surrounding surface features, (ii) the rate of change of one or more contours of the oval surface feature, (iii) indicators related to the stratigraphic unit containing the oval surface feature, or (iv) the specific contours of the oval feature.

[0083] Landform differentiation can exist in a variety of ways. As one example, because hydrogen toxicity kills many types of plant life, the presence or absence of plant evidence at the geographic location of an ovoid surface feature may indicate horizontal surface hydrogen infiltration, indicating active subsurface hydrogen accumulation somewhere in the nearby subsurface. As another example, in the case of bare soil, changes in surface features that are out of proportion to the surrounding topography may indicate elevated deformation rates, which, if present in association with the ovoid surface feature, may also indicate the possibility of active surface hydrogen infiltration. Exemplary embodiments may utilize DEMs with resolutions greater than about 10 m horizontally and about 5 m vertically to identify areas of landform differentiation from multispectral imagery, or LiDAR or InSAR. Surface features of interest typically form closed, ovoid-shaped surface depressions and thus have distinct three-dimensional shapes that can aid in identification.

[0084] As another method for assessing topographic differentiation, the relevance determination engine 214 may receive an image of the oval surface feature (from the memory 204, from another device via the communications circuitry 206, or from a user via the input-output circuitry 208). The relevance determination engine 214 may utilize one or more techniques to identify the degree of topographic differentiation between the oval surface feature and one or more surrounding surface features from the received image. For example, one or more object detection models trained by the model generator 210 and hosted by the image analysis engine 212 may segment an entire given image to detect surface features consistent with hydrogen accumulation as well as other surface features present in the given image. The relevance determination engine 214 may include one or more rule sets or statistical approaches for calculating the degree of topographic differentiation between the surface features consistent with hydrogen accumulation and the surrounding surface features.

[0085] The evaluation of the rate of change of one or more contours of an oval surface feature can be performed in many ways. For example, comparative analysis of multiple images can be used. To this end, the relevance determination engine 214 can receive two images of the oval surface feature captured at different times. These two images can be obtained via the communications circuitry 206 from separate devices (e.g., the remote sensing device 114, the user devices 118A-118N) or data stores, from the local memory 204, from a user via the input-output circuitry 208, or from a combination of these paths. Many remote sensing systems store historical image data going back years, and in some cases, this image data is of sufficient resolution to allow fine-grained comparative analysis to reveal changes in the contours of the oval surface at the centimeter scale. For oval surface features located in corresponding regions, this type of historical image data allows for automatic and immediate evaluation of images of the oval surface feature captured at different times to identify surface deformations or changes (or lack thereof) in the oval surface feature. Alternatively, if historical image data is unavailable, the relevance determination engine 214 may automatically prompt for the capture of new image data of the oval surface feature for future subsequent review and analysis of this type. The relevance determination engine 214 may then calculate the elapsed time between the capture of the two acquired images and the degree of difference in the segments of the two images that correspond to the oval surface feature. To this end, in some embodiments, it may be necessary to perform coregistration of the two images to ensure that the images are spatially aligned and allow for more accurate comparison. Coregistration aligns two images or elevation models in space. Image coregistration uses cross-correlation of pixel intensity patterns between images, while DEM coregistration uses similarities in surface relief. Both resolve shifts in image coordinates to align one image to another without modifying the underlying image or DEM information itself.After establishing that the two images are aligned, the relevance determination engine 214 may then evaluate changes between the images, such as a larger or smaller oval surface feature, changes in surface morphology (e.g., surface roughness or moisture content, as may be detected by SAR), or changes in the elevation profile of the oval surface feature (as may be received as a DEM optically generated by the hydrogen targeting system 102, LiDAR, or other data received from one or more remote sensing devices 114). After calculating both the time elapsed between the images and the degree of difference between the two images, the relevance determination engine 214 may then determine a rate of change in one or more contours of the oval surface feature.

[0086] Another mechanism for assessing the rate of change of one or more contours of the oval surface feature is to utilize seismic and / or other geophysical (e.g., gravity) data. To this end, the relevance determination engine 214 may first identify the geographic location of the oval surface feature. This may be done from evaluating various characteristics of the oval surface feature within the image based on metadata associated with the target image in which the oval surface feature is identified, or the geographic location may simply be received in operation 602 along with other information about the oval surface feature. The relevance determination engine 214 may then receive data indicative of past seismic (including microseismic) activity at the geographic location of the oval surface feature. This data may be obtained by the communications circuitry 206 from a separate device (e.g., remote sensing device 114, user devices 118A-118N) or data store (e.g., data store 106), from the local memory 204 or data store 106, from a user via the input-output circuitry 208, or from a combination of these paths. Existing seismic surveys, active seismometers, and geophones exist around the world that constantly monitor seismic activity, and this data is often publicly available. Depending on the specific location of the ovoid surface feature, there may be a historical record of seismic activity accurate enough to allow insight into subsurface activity at or near the ovoid surface feature. Thus, the relevance determination engine 214 may determine the rate of change of one or more contours of the ovoid surface based on past seismic activity at the geographic location of the ovoid surface feature.

[0087] Indications for stratigraphic units containing oval surface features may be sourced from publicly available or proprietary bed sections, well logs, and / or data store 106, or from various other resources related to the geography of the area containing the oval surface features. To this end, the relevance determination engine 214 may identify the geographic location of the oval surface feature and then receive information about the stratigraphic unit containing the identified geographic location. This information may be obtained by the communications circuitry 206 from a separate device (e.g., remote sensing device 114, user devices 118A-118N) or data store (e.g., data store 106), from local memory 204 or data store 106, from publicly available or proprietary bed sections, well logs, from a user via the input-output circuitry 208, or from a combination of these paths. Information about the stratigraphic unit may include (i) the geographic characteristics of the stratigraphic unit, (ii) hydrogen migration pathways to nearby reservoirs, traps, and seals, and the geographic location of ovoid surface features or other seeps, (iii) information about hydrogen traps or seals proximal to the geographic location of ovoid surface features, or (iv) the thermal maturity and / or current temperature of one or more portions of the stratigraphic unit.

[0088] Finally, the particular contour of an oval surface feature may affect the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation. To this end, as more oval surface features linked to hydrogen infiltration are identified, the model generator 210 may more accurately train the image classification model to identify unique features or contours associated with such oval surface features, as opposed to other types of oval surface features. Thus, the image analysis engine 212, including the image classification model, may be invoked by the relevance determination engine 214 to analyze the contour of the oval surface feature to determine the likelihood that the oval surface feature is caused by surface hydrogen infiltration rather than another geological event, where the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation includes the likelihood that the oval surface feature is caused by surface hydrogen infiltration rather than another geological event.

[0089] As indicated by operation 606, the apparatus 200 includes means, such as, for example, the relevance determination engine 214, for determining whether the estimated likelihood that the oval surface feature indicates active subsurface hydrogen accumulation satisfies a predetermined threshold. This predetermined threshold may be predefined and stored by the memory 204 of the apparatus 200 for retrieval during operation 606, or may be user-defined based on the risk appetite of the individual or entity initiating the analysis of the particular oval surface feature. Satisfaction of the predetermined threshold may be achieved when the estimated likelihood equals or exceeds a particular probability. It will be appreciated that in some embodiments, the relevance determination engine 214 may calculate the likelihood that the oval surface feature is estimated not to indicate active subsurface hydrogen accumulation, and in such embodiments, satisfaction of the predetermined threshold may actually be achieved when the relevance determination engine 214 estimates the likelihood to be below a particular probability. Based on whether the estimated likelihood satisfies the predetermined threshold, the procedure may proceed to either operation 608 (the predetermined threshold is not met) or operation 610 (the predetermined threshold is met).

[0090] As indicated by operation 608, apparatus 200 includes means, such as memory 204, communications circuitry 206, and input-output circuitry 208, for outputting an indication that the surface features do not indicate active subsurface hydrogen accumulation. Memory 204 may store the indication for subsequent use. Communications circuitry 206 may output the indication to a separate device (such as one of user devices 118A-118N) or may output the indication to a remote data store. Input-output circuitry 208 may deliver the indication directly to a user interacting with apparatus 200. In some embodiments, operation 608 may be optional, such that apparatus 200 may take no action if a predetermined threshold is not met. In such embodiments, apparatus 200 generates an output only when a predetermined threshold is met, thus limiting the output of data to scenarios in which high-value surface features are identified.

[0091] As indicated by operation 610, device 200 includes means, such as memory 204, communications circuitry 206, and input-output circuitry 208, for outputting an indication that the surface features indicate active subsurface hydrogen accumulation. Similar to operation 608, memory 204 may store the indication for subsequent use. Communications circuitry 206 may output the indication to a separate device (such as one of user devices 118A-118N) or may output the indication to a remote data store. Input-output circuitry 208 may deliver the indication directly to a user interacting with device 200, such as via a visualization layer overlaid on a map interface (e.g., Google Maps or GIS software). However, in some embodiments, device 200 may not take action at operation 610 if a predetermined threshold is met, in which case device 200 essentially functions as an alarm device that only indicates when high-value surface features are evaluated. In some embodiments, operation 610 may be optional, such that apparatus 200 may do nothing if a predetermined threshold is met. In such embodiments, apparatus 200 generates output only when the predetermined threshold is not met, thus limiting the output of data to scenarios where the value of the particular oval surface feature is low enough that it can be discarded.

[0092] It will be appreciated that while apparatus 200 may provide an indication of whether the surface feature indicates active subsurface hydrogen accumulation, optional operation 612 indicates that apparatus 200 may further include means, such as memory 204, communications circuitry 206, input-output circuitry 208, etc., for outputting an indication of the actual estimated likelihood that the oval surface feature indicates active subsurface hydrogen accumulation after estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation. This operation may include writing the indication to memory 204, transmitting the indication to a separate device (e.g., user devices 118A-118N depicted in FIG. 1 ), or delivering the indication to a user interacting with apparatus 200 via input-output circuitry 208. Apparatus 200 may output this indication before, after, or simultaneously with performance of operations 606, 608, or 610, or may output this indication as an alternative to performing operations 606, 608, or 610. In some embodiments, apparatus 200 may output this indicator in response to performance of operation 604 (as shown in FIG. 6 ), or in response to performance of either (or both) of operations 608 or 610. For example, the indicator of estimated likelihood may be withheld if a predetermined threshold is not met (e.g., if the procedure proceeds to operation 608, operation 612 is not called) because, in some implementations, knowledge of a particular estimated likelihood that a particular oval surface feature indicates active subsurface accumulation is irrelevant if the likelihood is low. Alternatively, in embodiments in which knowledge of a particular estimated likelihood for an oval surface feature is irrelevant if it has already been determined that the oval surface feature merits further attention, the indicator of estimated likelihood may be withheld if a predetermined threshold is met (e.g., if the procedure proceeds to operation 610, operation 612 is not called).

[0093] As described above, the illustrative embodiments provide methods and apparatus that enable improved identification of underground hydrogen accumulation. Given the recent and expected future growth in demand for hydrogen combined with the low carbon and / or energy intensity of current methods for producing man-made hydrogen, there is a large and growing need for tools that enable the production of natural hydrogen underground. The illustrative embodiments provide such tools that provide an automated, systematic, and comprehensive approach to identifying locations where hydrogen is likely to be produced, transported, and most importantly, stored (i.e., captured) within economical volumes underground.

[0094] 4, 5, and 6 illustrate operations performed by apparatuses, methods, and computer program products according to various exemplary embodiments. It will be understood that each flowchart block, and each combination of flowchart blocks, may be implemented by various means, embodied as hardware, firmware, circuitry, and / or other devices associated with the execution of software including one or more software instructions. For example, one or more of the operations described above may be embodied by software instructions. In this regard, software instructions embodying the above-described procedures may be stored by a memory of an apparatus employing embodiments of the present invention and executed by a processor of that apparatus. As will be understood, any such software instructions may be loaded into a computing device or other programmable apparatus (e.g., hardware) to generate a machine, such that the resulting computing device or other programmable apparatus implements the functions specified in the flowchart blocks. These software instructions may also be stored in a computer-readable memory that can direct the computing device or other programmable apparatus to function in a particular manner, such that the software instructions stored in the computer-readable memory generate an article of manufacture, the execution of which implements the functions specified in the flowchart blocks. The software instructions may also be loaded onto a computing device or other programmable apparatus and cause a sequence of operations to be performed on the computing device or other programmable apparatus to create a computer-implemented process, such that the software instructions executing on the computing device or other programmable apparatus provide operations to implement the functions specified in the flowchart blocks.

[0095] The flowchart blocks support combinations of means for performing the specified functions and combinations of acts for performing the specified functions. It will be understood that individual flowchart blocks and / or combinations of flowchart blocks may be implemented by special purpose hardware-based computing devices that perform particular functions, or by combinations of special purpose hardware and software instructions.

[0096] In some embodiments, some of the above operations may be modified or further amplified. Furthermore, in some embodiments, additional optional operations may be included. Modifications, amplifications, or additions to the above operations may be performed in any order and in any combination.

[0097] conclusion Many modifications and other embodiments of the inventions described herein will come to mind to one skilled in the art to which these inventions pertain having the benefit of the teachings presented in the foregoing description and the associated drawings. It is, therefore, to be understood that the invention is not limited to the particular embodiments disclosed, and that modifications and other embodiments are intended to be included within the scope of the appended claims. Furthermore, while the foregoing description and associated drawings describe example embodiments in the context of particular example combinations of elements and / or functions, it will be understood that alternative embodiments may provide different combinations of elements and / or functions without departing from the scope of the appended claims. In this regard, combinations of elements and / or functions other than those expressly described above are also contemplated, for example, as set forth in some of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.

[0098] (Appendix 1) 1. A method for training an image analysis engine to identify terrestrial surface features consistent with subsurface hydrogen accumulation, the method comprising: receiving, by a communications circuitry, a training dataset including a set of labeled images showing surface features consistent with subsurface hydrogen accumulation, the surface features consistent with subsurface hydrogen accumulation including oval surface depressions; training, by a model generator and using the training dataset, an image classification model of the image analysis engine to identify whether an image contains surface features consistent with subsurface hydrogen accumulation; hosting the trained image classification model by the image analysis engine.

[0099] (Appendix 2) receiving a target image by the communication circuitry; 2. The method of claim 1, further comprising: identifying, by the image analysis engine and using the image classification model, whether the target image contains any surface features consistent with subsurface hydrogen accumulation.

[0100] (Appendix 3) the image classification model includes an object detection model; 3. The method of claim 1 or 2, wherein training the image classification model further comprises training the image classification model to identify regions within images containing surface features consistent with subsurface hydrogen accumulation.

[0101] (Appendix 4) 4. The method of claim 3, wherein the training dataset includes a set of bounding boxes for each image in the set of labeled images, each bounding box for a particular image enclosing a corresponding segment of the particular image that includes surface features consistent with subsurface hydrogen accumulation.

[0102] (Appendix 5) the image classification model includes a semantic segmentation model; 5. The method of any one of claims 1-4, wherein training the image classification model includes training the image classification model to identify all pixels of the image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0103] (Appendix 6) 6. The method of claim 5, wherein the training dataset includes a set of pixel-wise masks for each image in the set of labeled images, the set of pixel-wise masks for a particular image collectively identifying whether all pixels of the particular image correspond to surface features consistent with subsurface hydrogen accumulation.

[0104] (Appendix 7) 7. The method of any one of claims 1 to 6, wherein the image classification model comprises a convolutional neural network.

[0105] (Appendix 8) The set of labeled images is Panchromatic, multispectral, or hyperspectral imagery, satellite imagery, or 8. The method of any one of claims 1 to 7, including panchromatic, multispectral, or hyperspectral satellite imagery.

[0106] (Appendix 9) 1. An apparatus for training an image analysis engine to identify aboveground surface features consistent with subsurface hydrogen accumulation, said apparatus comprising: a communication circuit configured to receive a training dataset including a set of labeled images exhibiting surface features consistent with subsurface hydrogen accumulation, the surface features consistent with subsurface hydrogen accumulation including oval surface depressions; and a model generator configured to use the training dataset to train an image classification model of the image analysis engine to identify whether an image contains surface features consistent with subsurface hydrogen accumulation; the image analysis engine is configured to host the trained image classification model.

[0107] (Appendix 10) the communication circuitry is further configured to receive an image of the target; 10. The apparatus of claim 9, wherein the image analysis engine is further configured to use the image classification model to identify whether the target image includes any surface features consistent with subsurface hydrogen accumulation.

[0108] (Appendix 11) the image classification model includes an object detection model; 11. The apparatus of claim 9 or 10, wherein the model generation unit is configured to train the image classification model to identify regions within images containing surface features consistent with subsurface hydrogen accumulation.

[0109] (Appendix 12) 12. The apparatus of claim 11, wherein the training dataset includes a set of bounding boxes for each image in the set of labeled images, each bounding box for a particular image enclosing a corresponding segment of the particular image that includes surface features consistent with subsurface hydrogen accumulation.

[0110] (Appendix 13) the image classification model includes a semantic segmentation model; 13. The apparatus of any one of claims 9-12, wherein the model generation device is configured to train the image classification model to identify all pixels of the image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0111] (Appendix 14) 14. The apparatus of claim 13, wherein the training dataset includes a set of pixel-wise masks for each image in the set of labeled images, the set of pixel-wise masks for a particular image collectively identifying whether all pixels of the particular image correspond to surface features consistent with subsurface hydrogen accumulation.

[0112] (Appendix 15) 15. The apparatus of any one of appendices 9 to 14, wherein the image classification model comprises a convolutional neural network.

[0113] (Appendix 16) The set of labeled images is Panchromatic, multispectral, or hyperspectral imagery, satellite imagery, or 16. The apparatus of any one of clauses 9-15, including panchromatic, multispectral, or hyperspectral satellite imagery.

[0114] (Appendix 17) 1. A computer program product for training an image analysis engine to identify aboveground surface features consistent with subsurface hydrogen accumulation, the computer program product comprising at least one non-transitory computer-readable storage medium storing software instructions that, when executed, cause an apparatus to: receiving a training dataset including a set of labeled images exhibiting surface features consistent with subsurface hydrogen accumulation, the surface features consistent with subsurface hydrogen accumulation including oval surface depressions; using the training dataset to train an image classification model of the image analysis engine to identify whether an image contains surface features consistent with subsurface hydrogen accumulation; a computer program product, wherein the trained image classification model is hosted by the image analysis engine;

[0115] (Appendix 18) The software instructions, when executed, cause the device to: receiving a target image; 18. The computer program product of claim 17, further comprising: using the image classification model to identify whether the target image includes any surface features consistent with subsurface hydrogen accumulation.

[0116] (Appendix 19) the image classification model includes an object detection model; 19. The computer program product of claim 17 or 18, wherein the software instructions, when executed, further cause the device to train the image classification model to identify regions within images containing surface features consistent with subsurface hydrogen accumulation.

[0117] (Appendix 20) 20. The computer program product of claim 19, wherein the training dataset includes a set of bounding boxes for each image in the set of labeled images, each bounding box for a particular image enclosing a corresponding segment of the particular image containing surface features consistent with subsurface hydrogen accumulation.

[0118] (Appendix 21) the image classification model includes a semantic segmentation model; 21. The computer program product of any one of claims 17-20, wherein the software instructions, when executed, further cause the device to train the image classification model to identify all pixels in the image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0119] (Appendix 22) 22. The computer program product of claim 21, wherein the training dataset includes a set of pixel-wise masks for each image in the set of labeled images, the set of pixel-wise masks for a particular image collectively identifying whether all pixels of the particular image correspond to surface features consistent with subsurface hydrogen accumulation.

[0120] (Appendix 23) 23. The computer program product of any one of Appendixes 17 to 22, wherein the image classification model comprises a convolutional neural network.

[0121] (Appendix 24) The set of labeled images is Panchromatic, multispectral, or hyperspectral imagery, satellite imagery, or 24. The computer program product of any one of Clauses 17-23, comprising panchromatic, multispectral, or hyperspectral satellite imagery.

[0122] (Appendix 25) 1. A method for automatically identifying terrestrial surface features consistent with subsurface hydrogen accumulation, the method comprising: receiving, by a communications circuit, a target image; identifying, by an image analysis engine and using a trained image classification model, whether the target image contains any surface features consistent with subsurface hydrogen accumulation; and outputting, by the communication circuitry, an indication of whether the target image includes any surface features consistent with subsurface hydrogen accumulation.

[0123] (Appendix 26) the trained image classification model comprises an object detection model; identifying whether the target image includes surface features consistent with subsurface hydrogen accumulation includes identifying any region within the target image that includes surface features consistent with subsurface hydrogen accumulation; 26. The method of claim 25, further comprising outputting, by the communications circuitry, any regions within the target image that contain surface features consistent with subsurface hydrogen accumulation.

[0124] (Appendix 27) the trained image classification model includes a semantic segmentation model; identifying whether the target image includes surface features consistent with subsurface hydrogen accumulation includes identifying all pixels in the target image that correspond to surface features consistent with subsurface hydrogen accumulation; 27. The method of claim 25 or 26, further comprising outputting, by the communications circuitry, an indication of all pixels in the target image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0125] (Appendix 28) 28. The method of any one of claims 25 to 27, further comprising performing an orthorectification on the target image by the image analysis engine to remove distortion before identifying whether the target image contains any surface features consistent with subsurface hydrogen accumulation.

[0126] (Appendix 29) 29. The method of any one of claims 25 to 28, wherein the surface features consistent with subsurface hydrogen accumulation include or lead to the creation of an oval surface depression.

[0127] (Appendix 30) 30. The method of any one of claims 25 to 29, wherein the trained image classification model comprises a convolutional neural network.

[0128] (Appendix 31) The target image is Panchromatic, multispectral, or hyperspectral imagery, satellite imagery, or 31. The method of any one of claims 25 to 30, including panchromatic, multispectral, or hyperspectral satellite imagery.

[0129] (Appendix 32) receiving, by the communication circuitry, a set of target images; identifying, by the image analysis engine and using the trained image classification model, all target images from the set of target images that contain any surface features consistent with subsurface hydrogen accumulation; 32. The method of any one of claims 25-31, further comprising outputting, by the communications circuitry, an indication of all target images from the set of target images that contain surface features consistent with subsurface hydrogen accumulation.

[0130] (Appendix 33) 1. An apparatus for automatically identifying terrestrial surface features consistent with subsurface hydrogen accumulation, said apparatus comprising: a communications circuit configured to receive an image of the target; an image analysis engine configured to use a trained image classification model to identify whether the target image contains any surface features consistent with subsurface hydrogen accumulation; The apparatus, wherein the communications circuitry is further configured to output an indication of whether the target image includes any surface features consistent with subsurface hydrogen accumulation.

[0131] (Appendix 34) the trained image classification model comprises an object detection model; the image analysis engine is configured to identify whether the target image includes surface features consistent with subsurface hydrogen accumulation by identifying any regions within the target image that include surface features consistent with subsurface hydrogen accumulation; 34. The apparatus of claim 33, wherein the communications circuitry is configured to output any regions within the target image that contain surface features consistent with subsurface hydrogen accumulation.

[0132] (Appendix 35) the trained image classification model includes a semantic segmentation model; the image analysis engine is configured to identify whether the target image includes surface features consistent with subsurface hydrogen accumulation by identifying all pixels in the target image that correspond to surface features consistent with subsurface hydrogen accumulation; 35. The apparatus of claim 33 or 34, wherein the communications circuitry is configured to output an indication of all pixels in the target image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0133] (Appendix 36) 36. The apparatus of any one of claims 33-35, wherein the image analysis engine is further configured to perform orthorectification on the target image to remove distortion before identifying whether the target image includes any surface features consistent with subsurface hydrogen accumulation.

[0134] (Appendix 37) 37. The apparatus of any one of claims 33-36, wherein the surface features consistent with subsurface hydrogen accumulation include or lead to the creation of an oval surface depression.

[0135] (Appendix 38) 38. The apparatus of any one of appendixes 33-37, wherein the trained image classification model comprises a convolutional neural network.

[0136] (Appendix 39) The target image is Panchromatic, multispectral, or hyperspectral imagery, satellite imagery, or 39. The apparatus of any one of clauses 33-38, including panchromatic, multispectral, or hyperspectral satellite imagery.

[0137] (Appendix 40) the communications circuitry is configured to receive a set of target images; the image analysis engine is configured to use the trained image classification model to identify all target images from the set of target images that include any surface features consistent with subsurface hydrogen accumulation; 40. The apparatus of any one of claims 33-39, wherein the communications circuitry is further configured to output an indication of all target images from the set of target images that include surface features consistent with subsurface hydrogen accumulation.

[0138] (Appendix 41) 1. A computer program product for automatically identifying aboveground surface features consistent with subsurface hydrogen accumulation, the computer program product comprising at least one non-transitory computer-readable storage medium storing software instructions that, when executed, cause an apparatus to: receiving a target image; using a trained image classification model to identify whether the target image contains any surface features consistent with subsurface hydrogen accumulation; and outputting an indication of whether the target image contains any surface features consistent with subsurface hydrogen accumulation.

[0139] (Appendix 42) the trained image classification model comprises an object detection model; the software instructions, when executed, cause the apparatus to identify whether the target image contains surface features consistent with subsurface hydrogen accumulation by identifying any regions within the target image that contain surface features consistent with subsurface hydrogen accumulation; 42. The computer program product of claim 41, wherein the software instructions, when executed, cause the device to output any regions within the target image that contain surface features consistent with subsurface hydrogen accumulation.

[0140] (Appendix 43) the trained image classification model includes a semantic segmentation model; the software instructions, when executed, cause the apparatus to identify whether the target image includes surface features consistent with subsurface hydrogen accumulation by identifying all pixels in the target image that correspond to surface features consistent with subsurface hydrogen accumulation; 43. The computer program product of claim 41 or 42, wherein the software instructions, when executed, cause the device to output an indication of all pixels in the target image that correspond to surface features consistent with subsurface hydrogen accumulation.

[0141] (Appendix 44) The software instructions, when executed, cause the device to: 44. The computer program product of any one of claims 41 to 43, further comprising performing an orthorectification on the target image to remove distortion before identifying whether the target image contains any surface features consistent with subsurface hydrogen accumulation.

[0142] (Appendix 45) 45. The computer program product of any one of claims 41 to 44, wherein the surface features consistent with subsurface hydrogen accumulation include or lead to the creation of an oval surface depression.

[0143] (Appendix 46) 46. The computer program product of any one of Clauses 41-45, wherein the trained image classification model comprises a convolutional neural network.

[0144] (Appendix 47) The target image is Panchromatic, multispectral, or hyperspectral imagery, satellite imagery, or 47. The computer program product of any one of Clauses 41-46, comprising panchromatic, multispectral, or hyperspectral satellite imagery.

[0145] (Appendix 48) The software instructions, when executed, cause the device to: receiving a set of target images; using the trained image classification model to identify all target images from the set of target images that contain any surface features consistent with subsurface hydrogen accumulation; and outputting an indication of all target images from the set of target images that contain surface features consistent with subsurface hydrogen accumulation.

[0146] (Appendix 49) 1. A method for automatically identifying terrestrial surface features indicative of active subsurface hydrogen accumulation, said method comprising: receiving, by communication circuitry, information describing said terrestrial oval surface feature; automatically estimating, by a relevance determination engine, the likelihood that said oval surface feature indicates active subsurface hydrogen accumulation; determining, by the relevance determination engine, whether the estimated likelihood meets a predetermined threshold; and in instances where the estimated likelihood meets the predetermined threshold, outputting, by the communication circuitry, an indication that the oval surface feature indicates active subsurface hydrogen accumulation.

[0147] (Appendix 50) automatically estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation; determining, by the association determination engine, a degree of topographic differentiation between the oval surface feature and one or more surrounding surface features; and estimating, by the relevance determination engine, the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the degree of topographic differentiation between the oval surface feature and the one or more surrounding surface features.

[0148] (Appendix 51) automatically estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation; determining, by the relevance determination engine, a rate of change of one or more contours of the oval surface feature; and estimating, by the relevance determination engine, the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the rate of change of the one or more contours of the oval surface feature.

[0149] (Appendix 52) Determining the rate of change of one or more contours of the oval surface feature comprises: obtaining, by said communication circuitry, two images of said oval surface feature captured at different times; calculating, by the relevance determination engine, the elapsed time between the capture of the two images; calculating, by the relevance determination engine, a degree of difference between the two images between the segments corresponding to the oval surface features; determining, by the relevance determination engine, the rate of change of the one or more contours of the oval surface feature based on the elapsed time between capture of the two images and the calculated degree of difference of the two images between the segments corresponding to the oval surface feature.

[0150] (Appendix 53) Determining the rate of change of one or more contours of the oval surface feature comprises: identifying, by the relevance determination engine, a geographic location of the oval surface feature; receiving, by the communications circuitry, data indicative of past microseismic activity at the geographic location of the oval surface feature; 52. The method of claim 51, comprising: determining, by the relevance determination engine, the rate of change of the one or more contours of the oval surface feature based on the data indicative of past microseismic activity at the geographic location of the oval surface feature.

[0151] (Appendix 54) automatically estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation; identifying, by the relevance determination engine, a geographic location of the oval surface feature; receiving, by the communications circuitry, information about a stratigraphic unit including the identified geographic location of the oval surface feature; and estimating, by the relevance determination engine, the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the received information about the stratigraphic unit.

[0152] (Appendix 55) The information about the stratigraphic units comprises: the geographic characteristics of said stratigraphic units; a hydrogen migration pathway to the geographic location of the oval surface feature; a hydrogen trap or seal proximate the geographic location of the oval surface feature; or 55. The method of claim 54, including at least one of thermal maturity or current temperature condition of one or more portions of the stratigraphic unit.

[0153] (Appendix 56) automatically estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation; analyzing, with an image analysis system, a contour of the oval surface feature to determine the likelihood that the oval surface feature is caused by surface hydrogen infiltration as opposed to another geographical event; 56. The method of any one of claims 49 to 55, wherein the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation comprises the likelihood that the oval surface feature is caused by surface hydrogen infiltration rather than another geological event.

[0154] (Appendix 57) 1. An apparatus for automatically identifying terrestrial surface features indicative of active subsurface hydrogen accumulation, said apparatus comprising: a communications circuit configured to receive information describing the terrestrial oval surface feature; 1. A relevance determination engine, comprising: automatically estimating the likelihood that the oval surface feature indicates an active subsurface hydrogen accumulation; and determining whether the estimated likelihood meets a predetermined threshold; and The apparatus, wherein the communications circuitry is further configured to output an indication that the oval surface feature indicates active subsurface hydrogen accumulation in instances where the estimated likelihood meets the predetermined threshold.

[0155] (Appendix 58) the relevance determination engine: determining a degree of topographic differentiation between the oval surface feature and one or more surrounding surface features; estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the degree of topographic differentiation between the oval surface feature and the one or more surrounding surface features.

[0156] (Appendix 59) the relevance determination engine: determining a rate of change of one or more contours of the oval surface feature; 59. The apparatus of claim 57 or 58, wherein the apparatus is configured to automatically estimate the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation by: estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the rate of change of the one or more contours of the oval surface feature.

[0157] (Appendix 60) the communication circuitry is further configured to obtain two images of the oval surface feature captured at different times; the relevance determination engine: calculating the elapsed time between the capture of the two images; calculating a measure of difference between the two images between segments corresponding to the oval surface feature; determining the rate of change of the one or more contours of the oval surface feature based on the elapsed time between capture of the two images and the calculated degree of difference of the two images between the segments corresponding to the oval surface feature.

[0158] (Appendix 61) the relevance determination engine is configured to identify a geographic location of the oval surface feature; the communications circuitry is configured to receive data indicative of past microseismic activity at the geographic location of the oval surface feature; 60. The apparatus of claim 59, wherein the relevance determination engine is configured to determine the rate of change of the one or more contours of the oval surface feature by determining the rate of change of the one or more contours of the oval surface feature based on the data indicative of past microseismic activity at the geographic location of the oval surface feature.

[0159] (Appendix 62) the relevance determination engine is configured to identify a geographic location of the oval surface feature; the communications circuitry is configured to receive information about a stratigraphic unit including the identified geographic location of the oval surface feature; 62. The apparatus of any one of claims 57 to 61, wherein the association determination engine is configured to automatically estimate the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation by estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the received information about the stratigraphic unit.

[0160] (Appendix 63) The information about the stratigraphic units comprises: the geographic characteristics of said stratigraphic units; a hydrogen migration pathway to the geographic location of the oval surface feature; a hydrogen trap or seal proximate the geographic location of the oval surface feature; or 63. The apparatus of claim 62, including at least one of thermal maturity or current temperature conditions of one or more portions of the stratigraphic unit.

[0161] (Appendix 64) an image analysis system configured to analyze a contour of the oval surface feature to determine a likelihood that the oval surface feature is caused by surface hydrogen infiltration rather than another geographic event; the relevance determination engine is configured to automatically estimate the likelihood that the oval surface feature indicates an active subsurface hydrogen accumulation by causing the image analysis system to analyze a contour of the oval surface feature to determine a likelihood that the oval surface feature is caused by surface hydrogen infiltration rather than another geological event; 64. The apparatus of any one of claims 57-63, wherein the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation includes the likelihood that the oval surface feature is caused by surface hydrogen infiltration rather than another geological event.

[0162] (Appendix 65) 1. A computer program product for automatically identifying terrestrial surface features indicative of active subsurface hydrogen accumulation, the computer program product comprising at least one non-transitory computer-readable storage medium storing software instructions that, when executed, cause an apparatus to: receiving information describing the terrestrial oval surface feature; automatically estimating the likelihood that the oval surface feature indicates an active subsurface hydrogen accumulation; and determining whether the estimated likelihood meets a predetermined threshold; and in instances where the estimated likelihood meets the predetermined threshold, outputting an indication that the surface feature is indicative of active subsurface hydrogen accumulation.

[0163] (Appendix 66) The software instructions, when executed, cause the device to: determining a degree of topographic differentiation between the oval surface feature and one or more surrounding surface features; and estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the degree of topographic differentiation between the oval surface feature and the one or more surrounding surface features.

[0164] (Appendix 67) The software instructions, when executed, cause the device to: determining, by a relevance determination engine, a rate of change of one or more contours of said oval surface feature; 67. The computer program product of claim 65 or 66, wherein the relevance determination engine automatically estimates the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation by: estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the rate of change of the one or more contours of the oval surface feature.

[0165] (Appendix 68) The software instructions, when executed, cause the device to: obtaining two images of the oval surface feature captured at different times; calculating the elapsed time between the capture of the two images; calculating a measure of difference between the two images between segments corresponding to the oval surface feature; determining the rate of change of the one or more contours of the oval surface feature based on the elapsed time between capture of the two images and the calculated degree of difference of the two images between the segments corresponding to the oval surface feature.

[0166] (Appendix 69) The software instructions, when executed, cause the device to: identifying a geographic location of the oval surface feature; receiving data indicative of past microseismic activity at the geographic location of the oval surface feature; and determining the rate of change of the one or more contours of the oval surface feature based on the data indicative of past microseismic activity at the geographic location of the oval surface feature.

[0167] (Appendix 70) The software instructions, when executed, cause the device to: identifying a geographic location of the oval surface feature; receiving information about a stratigraphic unit including the identified geographic location of the oval surface feature; 69. The computer program product of claim 65, further comprising: estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation based on the received information about the stratigraphic unit; and estimating the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation.

[0168] (Appendix 71) The information about the stratigraphic units comprises: the geographic characteristics of said stratigraphic units; a hydrogen migration pathway to the geographic location of the oval surface feature; a hydrogen trap or seal proximate the geographic location of the oval surface feature; or 71. The computer program product of claim 70, including at least one of thermal maturity or current temperature conditions of one or more portions of the stratigraphic unit.

[0169] (Appendix 72) The software instructions, when executed, cause the device to: automatically estimating the likelihood that the oval surface feature indicates an active subsurface hydrogen accumulation by analyzing a contour of the oval surface feature to determine the likelihood that the oval surface feature is caused by surface hydrogen infiltration rather than another geological event; 72. The computer program product of any one of claims 65 to 71, wherein the likelihood that the oval surface feature indicates active subsurface hydrogen accumulation comprises the likelihood that the oval surface feature is caused by surface hydrogen infiltration rather than another geological event.

[0170] (Appendix 73) 1. A method for automatically identifying terrestrial surface features indicative of active subsurface hydrogen accumulation, said method comprising: receiving, via a communications circuitry, a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation; training, by a model generator and using the training dataset, an object detection model to identify regions within the image containing surface features consistent with subsurface hydrogen accumulation; receiving a target image by the communication circuitry; identifying, by an image analysis engine and using the object detection model, regions within the target image that contain surface features consistent with subsurface hydrogen accumulation; automatically estimating, by a relevance determination engine, the likelihood that the surface features indicate active subsurface hydrogen accumulation; determining, by the relevance determination engine, whether the estimated likelihood meets a predetermined threshold; and in instances where the estimated likelihood meets the predetermined threshold, outputting, by the communication circuitry, an indication that the surface feature is indicative of active subsurface hydrogen accumulation.

[0171] (Appendix 74) the object detection model includes a semantic segmentation model; training the object detection model includes training the object detection model to identify all pixels of the image corresponding to surface features consistent with subsurface hydrogen accumulation; 74. The method of claim 73, wherein identifying regions within the target image including surface features consistent with subsurface hydrogen accumulation includes identifying all pixels within the target image that correspond to the surface features.

[0172] (Appendix 75) 75. The method of claim 73 or 74, further comprising performing an orthorectification on the target image to remove distortion before the image analysis engine identifies the region in the target image containing surface features consistent with subsurface hydrogen accumulation.

[0173] (Appendix 76) automatically estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation; determining, by the relevance determination engine, a degree of topographic differentiation between the surface feature and one or more surrounding surface features; and estimating, by the relevance determination engine, the likelihood that the surface feature indicates active subsurface hydrogen accumulation based on the degree of topographic differentiation between the surface feature and the one or more surrounding surface features.

[0174] (Appendix 77) automatically estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation; determining, by the relevance determination engine, a rate of change of one or more contours of the surface features; and estimating, by the relevance determination engine, the likelihood that the surface feature indicates active subsurface hydrogen accumulation based on the determined rate of change of the one or more contours of the surface feature.

[0175] (Appendix 78) automatically estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation; identifying, by the relevance determination engine, a geographic location of the surface feature; receiving, by the communications circuitry, information about stratigraphic units including the identified geographic locations of the surface features; and estimating, by the relevance determination engine, the likelihood that the surface feature indicates active subsurface hydrogen accumulation based on the received information about the stratigraphic unit.

[0176] (Appendix 79) The information about the stratigraphic units comprises: the geographic characteristics of said stratigraphic units; hydrogen migration pathways to the geographic locations of the surface features; a hydrogen trap or seal proximate the geographic location of the surface feature; or 79. The method of claim 78, including at least one of thermal maturity or current temperature condition of one or more portions of the stratigraphic unit.

[0177] (Appendix 80) 1. An apparatus for automatically identifying terrestrial surface features indicative of active subsurface hydrogen accumulation, said apparatus comprising: a communications circuit configured to receive a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation; a model generator configured to use the training dataset to train an object detection model to identify regions within an image containing surface features consistent with subsurface hydrogen accumulation, a model generator, the communications circuitry further configured to receive an image of the target; an image analysis engine configured to use the object detection model to identify regions within the target image that contain surface features consistent with subsurface hydrogen accumulation; 1. A relevance determination engine, comprising: automatically estimating the likelihood that the surface features indicate active subsurface hydrogen accumulation; determining whether the estimated likelihood meets a predetermined threshold; and The apparatus, wherein the communications circuitry is further configured to output an indication that the surface feature is indicative of active subsurface hydrogen accumulation in instances where the estimated likelihood meets the predetermined threshold.

[0178] (Appendix 81) the object detection model includes a semantic segmentation model; the model generator is configured to train the semantic segmentation model to identify all pixels of the image that correspond to surface features consistent with subsurface hydrogen accumulation; 81. The apparatus of claim 80, wherein the image analysis engine is configured to identify all pixels in the target image that correspond to the surface feature.

[0179] (Appendix 82) 82. The apparatus of claim 80 or 81, wherein the image analysis engine is further configured to perform orthorectification on the target image to remove distortion before identifying the region in the target image containing surface features consistent with subsurface hydrogen accumulation.

[0180] (Appendix 83) the relevance determination engine: determining a degree of topographic differentiation between the surface feature and one or more surrounding surface features; 83. The apparatus of any one of appendices 80-82, wherein the apparatus is configured to automatically estimate the likelihood that the surface feature indicates active subsurface hydrogen accumulation by: estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation based on the degree of topographic differentiation between the surface feature and the one or more surrounding surface features.

[0181] (Appendix 84) the relevance determination engine: determining a rate of change of one or more contours of said surface features; 84. The apparatus of any one of appendices 80 to 83, wherein the apparatus is configured to automatically estimate the likelihood that the surface feature indicates active subsurface hydrogen accumulation by: estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation based on the determined rate of change of the one or more contours of the surface feature.

[0182] (Appendix 85) the relevance determination engine is further configured to identify a geographic location of the surface feature; the communications circuitry is further configured to receive information about stratigraphic units that include the identified geographic locations of the surface features; 85. The apparatus of any one of claims 80 to 84, wherein the relevance determination engine is configured to automatically estimate the likelihood that the surface feature indicates active subsurface hydrogen accumulation by estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation based on the received information regarding the stratigraphic unit.

[0183] (Appendix 86) The information about the stratigraphic units comprises: the geographic characteristics of said stratigraphic units; hydrogen migration pathways to the geographic locations of the surface features; a hydrogen trap or seal proximate the geographic location of the surface feature; or 86. The apparatus of claim 85, including at least one of thermal maturity or current temperature conditions of one or more portions of the stratigraphic unit.

[0184] (Appendix 87) 1. A computer program product for automatically identifying terrestrial surface features indicative of active subsurface hydrogen accumulation, the computer program product comprising at least one non-transitory computer-readable storage medium storing software instructions that, when executed, cause an apparatus to: receiving a training dataset of labeled images showing surface features consistent with subsurface hydrogen accumulation; using the training dataset to train an object detection model to identify regions within the image containing surface features consistent with subsurface hydrogen accumulation; receiving a target image; identifying regions within the target image containing surface features consistent with subsurface hydrogen accumulation; automatically estimating the likelihood that the surface features indicate active subsurface hydrogen accumulation; determining whether the estimated likelihood meets a predetermined threshold; and in instances where the estimated likelihood meets the predetermined threshold, outputting an indication that the surface feature is indicative of active subsurface hydrogen accumulation.

[0185] (Appendix 88) the object detection model includes a semantic segmentation model; the software instructions, when executed, cause the apparatus to train the object detection model to identify all pixels of an image that correspond to surface features consistent with subsurface hydrogen accumulation; 88. The computer program product of claim 87, wherein the software instructions, when executed, cause the device to identify regions within the target image containing surface features consistent with subsurface hydrogen accumulation by identifying all pixels within the target image that correspond to the surface features.

[0186] (Appendix 89) The software instructions, when executed, cause the device to: 89. The computer program product of claim 87 or 88, further comprising performing an orthorectification on the target image to remove distortion before identifying the region in the target image containing surface features consistent with the subsurface hydrogen accumulation.

[0187] (Appendix 90) The software instructions, when executed, cause the device to: determining a degree of topographic differentiation between the surface feature and one or more surrounding surface features; 90. The computer program product of any one of appendices 87-89, further comprising: estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation based on the degree of topographic differentiation between the surface feature and the one or more surrounding surface features; and

[0188] (Appendix 91) The software instructions, when executed, cause the device to: determining a rate of change of one or more contours of said surface features; 91. The computer program product of any one of appendices 87 to 90, further comprising: estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation based on the determined rate of change of the one or more contours of the surface feature; and estimating the likelihood that the surface feature indicates active subsurface hydrogen accumulation.

[0189] (Appendix 92) The software instructions, when executed, cause the device to: identifying the geographic location of the surface features; receiving information about a stratigraphic unit including the identified geographic location of the surface feature; 92. The computer program product of any one of appendices 87 to 91, further comprising: estimating the likelihood that the surface features indicate active subsurface hydrogen accumulation based on the received information about the stratigraphic unit; and estimating the likelihood that the surface features indicate active subsurface hydrogen accumulation.

[0190] (Appendix 93) The information about the stratigraphic units comprises: the geographic characteristics of said stratigraphic units; hydrogen migration pathways to the geographic locations of the surface features; a hydrogen trap or seal proximate the geographic location of the surface feature; or 93. The computer program product of claim 92, including at least one of thermal maturity or current temperature conditions of one or more portions of the stratigraphic unit.

Claims

[Claim 1] 1. A method for training an image analysis engine to identify terrestrial surface features consistent with subsurface hydrogen accumulation, the method comprising: receiving, by a communications circuitry, a training dataset including a set of labeled images showing surface features consistent with subsurface hydrogen accumulation, the surface features consistent with subsurface hydrogen accumulation including oval surface depressions; training, by a model generator and using the training dataset, an image classification model of the image analysis engine to identify whether an image contains surface features consistent with subsurface hydrogen accumulation; hosting the trained image classification model by the image analysis engine.

Images